def plot_from_exps(
exp_data,
filters={},
split_figures_by=None,
split_plots_by=None,
x_key='n_timesteps',
y_key=None,
sup_y_key=None,
plot_name='./bad-models.png',
subfigure_titles=None,
plot_labels=None,
x_label=None,
y_label=None,
num_rows=1,
y_limits=None,
report_max_performance=False,
log_scale=False,
round_x=None,
):
exp_data = filter(exp_data, filters=filters)
exps_per_plot = group_by(exp_data, group_by_key=split_figures_by)
num_columns = len(exps_per_plot.keys())
assert num_columns % num_rows == 0
num_columns = num_columns // num_rows
fig, axarr = plt.subplots(num_rows, num_columns, figsize=(14, 8))
axarr = np.reshape(axarr, (num_rows, num_columns))
fig.tight_layout(pad=5.0, w_pad=1, h_pad=2, rect=[0, 0, 1, 1])
# iterate over subfigures
for i, (default_plot_title,
plot_exps) in enumerate(sorted(exps_per_plot.items())):
plots_in_figure_exps = group_by(plot_exps, split_plots_by)
subfigure_title = subfigure_titles[
i] if subfigure_titles else default_plot_title
r, c = i // num_columns, i % num_columns
axarr[r, c].set_title(subfigure_title)
axarr[r, c].xaxis.set_major_locator(plt.MaxNLocator(5))
axarr[r, c].ticklabel_format(style='sci', axis='x', scilimits=(0, 0))
# iterate over plots in figure
y_max_mean = -1e10
y_axis_min = 1e10
y_axis_max = -1e10
for j, default_label in enumerate(
sorted(plots_in_figure_exps, key=sorting_legend)):
exps = plots_in_figure_exps[default_label]
x, y_mean, y_std, x_limits = prepare_data_for_plot(
exps, y_key=y_key, sup_y_key=sup_y_key, round_x=round_x)
label = plot_labels[j] if plot_labels else default_label
_label = label if i == 0 else "__nolabel__"
if log_scale:
axarr[r, c].semilogx(x,
y_mean,
label=_label,
linewidth=LINEWIDTH,
color=get_color(label))
else:
axarr[r, c].plot(x,
y_mean,
label=_label,
linewidth=LINEWIDTH,
color=get_color(label))
axarr[r, c].fill_between(x,
y_mean + y_std,
y_mean - y_std,
alpha=0.2,
color=get_color(label))
# axis labels
axarr[r, c].set_xlabel(x_label if x_label else x_key)
if c == 0:
axarr[r, c].set_ylabel(y_label if y_label else y_key)
if x_limits is not None:
axarr[r, c].set_xlim(*x_limits)
if y_limits is not None:
axarr[r, c].set_ylim(*y_limits)
else:
_y_axis_min, _y_axis_max = correct_limit(
axarr[r, c], x, y_mean - y_std, y_mean + y_std)
y_axis_max = max(_y_axis_max, y_axis_max)
y_axis_min = min(_y_axis_min, y_axis_min)
if max(y_mean) > y_max_mean:
y_max_mean = max(y_mean)
if report_max_performance:
label = 'max' if i == 0 else "__nolabel__"
axarr[r, c].plot(axarr[r, c].get_xlim(), [y_max_mean] * 2,
'k--',
label=label)
if y_limits is None:
axarr[r, c].set_ylim([y_axis_min, y_axis_max])
fig.legend(loc='lower center',
ncol=4,
bbox_transform=plt.gcf().transFigure)
fig.savefig(plot_name)
axarr[r, c].plot(axarr[r, c].get_xlim(), [y_max_mean] * 2,
'k--',
label=label)
if y_limits is None:
axarr[r, c].set_ylim([y_axis_min, y_axis_max])
fig.legend(loc='lower center',
ncol=3,
bbox_transform=plt.gcf().transFigure)
fig.savefig(plot_name)
# filter_dict = {'fast_lr': 0.001, 'path_length':200}
filter_dict = {}
exps_data_filtered = filter(exps_data, filter_dict)
plot_from_exps(
exps_data,
split_figures_by='env.$class',
split_plots_by='algo',
y_key='train-AverageReturn',
x_key='Time',
filters=filter_dict,
sup_y_key=[
'EnvTrajs-AverageReturn', 'Data-EnvTrajs-AverageReturn',
'EnvTrajs-AverageReturn', 'train-AverageReturn'
],
# subfigure_titles=['HalfCheetah - output_bias_range [0.0, 0.1]',
# 'HalfCheetah - output_bias_range [0.0, 0.5]',
# 'HalfCheetah - output_bias_range [0.0, 1.0]'],